University Of Illinoisfaculty.las.illinois.edu/swillia3/www/Econ 557/k-double auction... ·...
Transcript of University Of Illinoisfaculty.las.illinois.edu/swillia3/www/Econ 557/k-double auction... ·...
1. Introduction
(a) double auction: any procedure in which buyers and sellers interact
to arrange trade.
i. contrast with the passive role of the seller in most auction models
ii. can be dynamic or one shot; most theory focuses on one-shot
procedures, but most experimental work has focused on dynamic
procedures
(b) What are the main issues and motivations?
i. experimental evidence dating from the 1960s (V. Smith, inspired
by E. Chamberlin’s classroom experiments at Harvard)
A. "clearinghouse" (one-shot) versus continuous time
ii. price discovery vs. price verification (R. Wilson, Reny and Perry
(2006)): where do prices come from?
iii. M. A. Satterthwaite: strategic behavior is a problem in small
markets but not in large markets, and it doesn’t require a lot of
traders for a market to be large
A. Myerson-Satterthwaite
B. Gibbard-Satterthwaite
iv. market design
A. design of algorithms for computerized trading
B. in parallel to the use of auction theory to inform the design
of auctions
C. Budish, Cramton and Shim (2014)
D. Loertscher and Mazzetti (2014), "A Prior-Free Approximately
Optimal Dominant-Strategy Double Auction"
2. Elementary model of a static double auction: independent, private values
(a) Chatterjee and Samuelson (1983) in the bilateral case
(b) buyers, each of whom wishes to buy at most one unit of an in-
divisible good, sellers, each of whom has one unit of the good to
sell
(c) redemption value/cost: ∈ [ ], v , ∈ [ ], v . Here,
for simplicity, [ ]=[ ] = [0 1]
(d) for ∈ [0 1], -double auction in bilateral case:i. bid , ask
ii. trade iff ≥ at price + (1− )
iii. = 1: buyer’s bid double auction
A. dominant strategy of seller to submit his cost as his ask
iv. ∈ (0 1)
1
v. FOCs, assuming increasing strategies:
( ) =
Z −1()
0
( − (+ (1− )()))()
( ) = ( − ¡+ (1− )(−1 ())¢(−1 ()) · 1
0 (−1 ())−
¡−1 ()
¢= ( − ) · (−1 ())
0 (−1 ())−
¡−1 ()
¢
( ) =
Z 1
−1()((() + (1− ))− )()
( ) = −(¡(−1()) + (1− )¢− )(−1())
· 1
0 (−1())+ (1− )
¡1−
¡−1 ()
¢¢= −(− ) · (−1())
0 (−1())+ (1− )
¡1−
¡−1 ()
¢¢equilibrium:
0 = ( −()) · (−1())0 (−1())
− ¡−1 (())
¢= −(()− ) · (−1(())
0 (−1(())+ (1− )
¡1−(−1()
¢vi. "linked" differential equations
vii. Chatterjee-Samuelson linear solution in uniform case:
() =
½23 + 1
12if ≥ 1
4
if ≥ 14
() =
½23+ 1
4if ≤ 3
4
if ≥ 34
viii. in general: every system of differential equations has a geometric
representation
0 ≤ ≤ ≤ 1
0 = ( − ) · ()̇− ()
0 = −(− ) · ()̇ + (1− ) (1− ())
ix. Increasing strategies:
A. probability of trading must be nondecreasing in a buyer’s
value and nonincreasing in a seller’s cost
B. "no flat spots" in the multilateral case
2
x. Sufficiency of FOC: Sufficiency of FOC
evaluating the buyer’s FOC at bid , value −1():
0 = (−1 ()− ) · (−1()0 (−1())
− ¡−1 ()
¢⇔(−1 ()− ) =
¡−1 ()
¢0 (−1())
0¡−1()
¢marginal utility with value and bid :
( ) = ( − ) · (−1 ())0 (−1 ())
− ¡−1 ()
¢=
(−1 ())0 (−1 ())
"( − )−
¡−1 ()
¢(−1 ())
0¡−1 ()
¢#
=(−1 ())0 (−1 ())
£( − )− ¡−1()−
¢¤=
(−1 ())0 (−1 ())
£( −−1())
¤xi. existence of "double continuum" of equilibria
3
A. Leininger, Linhart and Radner (1989): step function equilib-
ria
B. Linhart and Radner (1989) : minmax regret and minmax
expected regret
(e) Multilateral case: bids, asks. Expressing market-clearing price
in terms of order statistics.
(1) ≤ (2) ≤ ≤ (+)
Assuming () (+1):
bids asks
≥ (+1)
≤ () −
4
Figure 1:
In the case of () = (+1): + 1
bids asks
() = (+1) − −
= () = (+1)
() = (+1) − −
We have + (− − ) ⇒ + ; there are enough units for
sale at = () = (+1) to satisfy every buyer who is willing to pay
more than this price. Similarly, + ( − − ) ⇒ + ,
and so there are enough buyers willing to buy at this price to satisfy
every seller who is willing to accept less than this price. At issue is
allocating among the + traders who bid/ask = () = (+1).
If + + , then there is excess demand at the price = () =
(+1); satisfy all of the buyers who bid more than the price and then
randomly allocate the remaining supply among those buyers whose
bids equaled the price. If + + , then we have excess supply
at the price. Allow all the sellers whose asks were below to sell and
then randomly choose among those whose asks equaled the price to
determine who gets to sell.
(f) no trade equilibrium
(g) the BBDA: = (+1) with sellers trading only if their asks are
strictly less than the price
5
i. FOC:
( ) = ( − ) Pr( = ())− Pr(() (+1))
Pr(() (+1)) =
−1X=0
µ− 1
¶µ
−
¶¡−1()
¢·¡1−¡−1()
¢¢−1− ()
−(1− ())
−+
Pr( = ()) =
()·−1X=0
µ− 1
¶µ− 1
− 1−
¶¡−1()
¢ ¡1−
¡−1()
¢¢−1− · ()
−1−(1− ())
−+
+(−1) (−1())
0(−1())·−1X=0
µ− 2
¶µ− 2
− 1−
¶¡−1()
¢ ¡1−
¡−1()
¢¢−2− · ()
−1−(1− ())
−++1
ii. existence of equilibrium
6
iii. uniform on [0 1]: ()= +1
iv. k-DA
7
1. (a) Convergence results: , satisfy the bounds
1
≤
≤
8
for some constant . Let denote an equilibrium in the
market with buyers and sellers.
i. Convergence to price-taking behavior at the rate (1): There
exists a constant 1 () such that
−() ()− ≤ 1 ()
ii. Convergence to efficiency:
A. relative inefficiency:
−
B. relative inefficiency is (12), i.e., there exists a constant
2 () such that
−
≤ 2 ()
2
iii. meaningfulness of rates of convergence: statistics
A. numerical results
(b) Are these rates fast? M. A. Satterthwaite and S. R. Williams, "The
Optimality of a Simple Market Mechanism". Econometrica, Vol. 70,
No. 5 (Sep., 2002), pp. 1841-1863.
i. the k-DA is worst case asymptotic optimal among all Bayesian
incentive compatible, interim individually rational and ex ante
budget balanced mechanisms
A. asymptotic: mechanisms are compared using the rates at
which relative inefficiency converges to zero
B. worst-case: each mechanism is evaluated in its least favorable
environment (i.e., the choice of the distributions and )
9
ii. Result
A. constrained efficient mechanism in the sense of Myerson
B. relative inefficiency of the constrained efficient mechanism is
at least 2 in the uniform case for some
C. We believe that this holds for all , that are reasonably
well-behaved
D. relative inefficiency of any mechanism in its worst case ≥relative inefficiency of constrained efficient mechanism in its
worst case ≥ 2
E. Applying a term from computer science, the main result of
this paper is that the k-DA is worst-case asymptotic optimal
among all mechanisms for organizing trade that satisfy these
two constraints. "Asymptotic" refers here to the ranking
of mechanisms using rates of convergence and "worst-case"
refers to the evaluation of each mechanism in its least favor-
able environment for each value of m. Stated simply, this
result means that the k-DA’s worst-case error over a set of
environments converges to zero at the fastest possible rate
among all interim individually rational and ex ante budget
balanced mechanisms.
2. Related Work
(a) T. A. Gresik and M. A. Satterthwaite, "The Rate at Which a Simple
Market Converges to Efficiency as the Number of Traders Increases:
An Asymptotic Result for Optimal Trading Mechanisms", J. of Econ.
Theory 48 (1989), pp. 304-332.
i. Along with Wilson (1985), these two papers are to first to go be-
yond the bilateral bargaining model of Chatterjee and Samuelson
(1983) to the multilateral case.
ii. markets with buyers and sellers for ∈ Niii. relative inefficiency in the constrained efficient mechanism is
³ln 2
´,
a result that has been superceded by the rate established for the
-DA
iv. notable for the question that it pursues, which is the origin of all
of the convergence results that I discuss above
(b) R. Wilson, "Incentive Efficiency of Double Auctions". Econometrica,
Vol. 53, No. 5 (Sep., 1985), pp. 1101-1115.
i. If min {} is sufficiently large, then an equilibrium of a -
double auction is interim incentive efficient in the sense of Holm-
ström and Myerson (1983)
A. It cannot be common knowledge at the interim stage that
some other equilibrium of some possibly different procedure
10
Pareto dominates the equilibrium of the -double auction
under consideration
B. endurance of simple procedures such as the -double auction
ii. assumes existence of increasing, differentiable equilibrium strate-
gies with derivatives bounded uniformly for all ,
iii. imputes welfare weights from the allocation rule in the -DA in
equilibrium
A. the issue is whether or not these imputed weights are positive
as required for interim incentive efficiency
B. these imputed weights converge uniformly for all trader types
to 1 as min {} → ∞, and so for large min {} theyare positive for all trader types
C. no intuition as to why a large market is required for interim
incentive efficiency; market size is a means to an end
D. no examples of small markets in which an equilibrium fails to
be interim incentive efficient, except the no-trade equilibrium
iv. two aspects of the Wilson Critique:
A. procedures that are not defined in terms of the probabilistic
beliefs of the agents (traders)
B. relaxing the assumption of common knowledge of beliefs
C. p. 1114: "The practical advantage of a double auction is
that its rules for trades and payments do not invoke the data
that are common knowledge among the agents — namely, the
numbers of buyers and sellers, the joint probability distrib-
ution of their types, and the functional dependence of their
reservation prices on the type parameter. Instead, the bur-
den of coping with the complexity of the common knowledge
features is assumed by the traders in the construction of their
strategies."
(c) P. McAfee, "A Dominant Strategy Double Auction." Journal of Eco-
nomic Theory, Vol. 56, No. 2 (April 1992), pp. 434-450.
i. When ((+1) + (+1))2 ∈ [() ()], = ((+1) + (+1))2
and trades are made.
ii. When ((+1) + (+1))2 ∈ [() ()], highest − 1 buyers pay(), lowest − 1 sellers receive (), − 1 trades made andmonetary surplus of ( − 1)(() − ())
iii.
11
iv. virtues:
A. dominant strategies
B. if the monetary surplus of the "specialist" is counted among
the gains from trade, then expected inefficiency is (1 (+ ))
v. flaw: does not converge to efficiency if the monetary surplus is
treated as a cost of trading to the traders
vi. Loertscher and Marx (2015)
(d) Large Double Auctions
i. approach of my work with Satterthwaite
A. focus on the first order conditions
B. analyzed using combinatorics
ii. large double auctions: assumes a sufficiently large number of
traders
A. results of probability and statistics become applicable
B. asymptotics
C. remains a model of strategic price discovery
D. rarely the production of an equilibrium or any connection to
smaller markets
E. motivation typically based upon longstanding problems in
microeconomic theory: strategic foundation for competitive
equilibrium and for REE
12
iii. M. W. Cripps and J. M. Swinkels, "Efficiency of Large Double
Auctions". Econometrica, Vol. 74, No. 1 (Jan., 2006), pp. 47-
92.
A. result: As the number fo traders grows every nontrivial equi-
librium of the double auction setting converges to the Wal-
rasian outcome. Relative inefficiency disappears at the rate
12− for any 0
B. two goods, initially unitary supply/demand, later multiple
units are allowed
C. considers sequences of markets with possibly correlated, pri-
vate values in [0 1]
D. ≤symmetry of the distribution and across the strategies usedby each side of the market is allowed
E. Cripps and Swinkels: symmetry of strategies assumes away
the problem of making sure that the units are allocated prop-
erly to each side of the market
F. restriction on distribution: no asymptotic gaps, no asymp-
totic atoms
G. for ∈ (0 1], z-independence: bounds the amount that thedistribution of any trader’s value changes conditional on the
values of the remaining traders
H. asymmetry and "purification" of equilibrium strategies as the
number of traders grows
I. to establish the rate of convergence, "quite large" is neces-
sary. No indication of when the rate begins to be observed.
iv. P. J. Reny and M. Perry, "Toward a Strategic Foundation for
Rational Expectations Equilibrium", Econometrica, Vol. 74, No.
5 (Sep., 2006), pp. 1231-1269.
A. provide a strategic foundation for rational expectations equi-
librium
B. affiliated, interdependent values/costs
C. limit market, with a continuum of traders and real-valued
bids/asks: Bayes-Nash equilibrium in increasing strategies
that implements a fully-revealing REE price
D. main result is continuity as the number of traders and the
number of possible bids/asks goes to infinity: for a suffi-
ciently large number of traders, and for a discrete grid of
possible bids/asks that is sufficiently fine, there exists a BNE
in increasing strategies that approximates the equilibrium of
the limit market
E. all traders are fully rational and strategic: no noise traders
and sellers are active (unlike auction models)
F. No indication of how large a market is required, no examples
in finite markets
13
G. "Our main insight is that establishing the existence of a
monotone equilibrium poses serious difficulties only when in-
dividual agents can have a significant impact on the price."
Is this a problem of "proof", or truly of existence? They
suggest the later.
(e) R. C. Shafer, "Convergence to Price-Taking by Regret-Minimizers in
-Double Auctions."
i. P. B. Linhart and R. Radner, "Minimax-Regret Strategies for
Bargaining over Several Variables". J. of Econ. Theory 48, 152-
178 (1989).
ii. Shafer: Alternatives to Bayesian decision-making in modeling
how traders select their bids and asks in a -DA
iii. Does the emergence of price-taking behavior as the market in-
creases in size fundamentally require that one be a Bayesian?
iv. minimax regret and maxmin: behavior invariant to the size of
the market
v. culprit: this is true of any decision rule that satisfies the axiom
of symmetry
vi. Γ-minimax regret, and Γ-maxmin; minimizing maximum expected
regret
(f) J. H. Kagel andW. Vogt, "Buyer’s Bid Double Auctions: Preliminary
Experimental Results". Chapter 10 in The Double Auction Market:
Institutions, Theories and Evidence, D. Friedman and J. Rust, eds.
Addison Wesley (1993): Reading.
i. experimental design
A. = 2 and = 8 traders on each side
ii. few sellers played their dominant strategies, causing inefficiency
iii. buyers underbid by less than the equilibrium prediction
iv. change from = 2 to = 8 notable but not as much as pre-
dicted by theory
A. this is largely attributable to the fact that the underbidding
by buyers in the case of = 2 is not as extreme as theory
predicts, leaving little room for improvement
v. opportunities for learning in BBDA
(g) Continuous Bid/Ask Market
i. R. Wilson, "Equilibria of Bid-Ask Markets," in: Arrow and the
Ascent of Economic Theory: Essays in Honor of Kenneth J.
Arrow, G. Feiwel (ed.); Chapter 11, pp. 375-414. London and
New York: Macmillan Press and New York University Press,
1987.
14
ii. D. Easley and J. Ledyard, "Theories of Price Formation and
Exchange in Double Oral Auctions," in: The Double Auction
Market: Institutions, Theories and Evidence, D. Friedman and
J. Rust, eds. Addison Wesley (1993): Reading.
iii. D. Friedman, "A Simple Testable Model of Double Auction Mar-
kets". J. Econ. Behav. & Org. (1991), 47-70.
iv. D. Friedman, "The Double Auction Market Institution: A Sur-
vey", in: The Double Auction Market: Institutions, Theories
and Evidence, D. Friedman and J. Rust, eds. Addison Wesley
(1993): Reading.
3. Asymptotics
(a) goals
i. identify the asymptotic distribution of the BBDA’s price
ii. identify the asymptotic limits of the probabilities in a trader’s
FOC
iii. formulate the asymptotic FOCs (AFOCs) and solve
iv. compare the solutions to the AFOCs to computed equilibrium
v. AFOCs identify what is "first order" in a trader’s decision prob-
lem
A. comparative statics in the distributions and the numbers of
traders
(b) review
i. CPV environment
ii. informational model
iii. convergence results
iv. limit market: Let ≡ (+) and (ksi) ≡ −1 (), the th
quantile of distribution . The limit market in state consists
of probability masses of measure of buyers and measure 1−
of sellers with values/costs , which conditional on are i.i.d.
according to ( − ).
v. REE: The unique REE price in the limit market in state is
REE ≡ +
(c) fix , ; markets with buyers and sellers
i. () = :(+)−1, () = +1:(+)−1, () = +1:(+)
ii. (), () and e() are asymptotically consistent, unbiased and
normal estimators of the REE price in state :
() () ∼ ANµREE
(+ )2
[(+ )− 1] 2()¶
e() ∼ ANµREE
(+ )2
(+ )2()
¶
15
iii. figures that depict the distribution of e()
(d) FOC:
( − )()|(|)− Pr[() ≤ ≤ ()|] = 0i. asymptotic offset:
approx () =1
(+ ) − 11
()
A. prices centered on REE ≡ + B. no distinction between and except in determining
C. dependence on ()
(e) Numerical Example
i.
16
A. note: comparison between table
B. accuracy of approximation
ii.
17